The Effect of Internet Security Breach Announcements On Market Value: Capital Market Reactions For Breached Firms and Internet Security Developers

International Journal of Electronic Commerce
ISSN: 1086-4415 (Print) 1557-9301 (Online) Journal homepage: http://www.tandfonline.com/loi/mjec20
The Effect of Internet Security Breach

Announcements on Market Value: Capital Market
Reactions for Breached Firms and Internet
Security Developers
Huseyin Cavusoglu Ph.D. , Birendra Mishra Ph.D. & Srinivasan Raghunathan

Ph.D.
To cite this article: Huseyin Cavusoglu Ph.D. , Birendra Mishra Ph.D. & Srinivasan Raghunathan
Ph.D. (2004) The Effect of Internet Security Breach Announcements on Market Value: Capital
Market Reactions for Breached Firms and Internet Security Developers, International Journal of
Electronic Commerce, 9:1, 70-104
To link to this article: http://dx.doi.org/10.1080/10864415.2004.11044320
Published online: 08 Dec 2014.
Submit your article to this journal
Article views: 38
View related articles
Citing articles: 4 View citing articles
Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=mjec20
Download by: [University of Nebraska, Lincoln] Date: 09 October 2015, At: 01:25
The Effect of Internet Security Breach
Announcements on Market Value:
Capital Market Reactions for Breached Firms
and Internet Security Developers
Huseyin Cavusoglu, Birendra Mishra, and Srinivasan Raghunathan
ABSTRACT: Assessing the value of information technology (IT) security is challenging

because of the difficulty of measuring the cost of security breaches. An event-study analysis,
using market valuations, was used to assess the impact of security breaches on the market
Downloaded by [University of Nebraska, Lincoln] at 01:25 09 October 2015
value of breached firms. The information-transfer effect of security breaches (i.e., their
effect on the market value of firms that develop security technology) was also studied.
The results show that announcing an Internet security breach is negatively associated
with the market value of the announcing firm. The breached firms in the sample lost, on
average, 2.1 percent of their market value within two days of the announcement—an
average loss in market capitalization of $1.65 billion per breach. Firm type, firm size, and
the year the breach occurred help explain the cross-sectional variations in abnormal returns
produced by security breaches. The effects of security breaches are not restricted to the
breached firms. The market value of security developers is positively associated with the
disclosure of security breaches by other firms. The security developers in the sample realized
an average abnormal return of 1.36 percent during the two-day period after the
announcement—an average gain of $1.06 billion in two days. The study suggests that the
cost of poor security is very high for investors.
KEY WORDS AND PHRASES: Capital markets, event study, information technology secu-
rity, information technology security management, Internet security, security breach an-
nouncements.
The number of companies that conduct business over the Internet is steadily
increasing, but the massive growth of e-business has not been an unmitigated
boon. The Internet provides great advantages to firms, facilitating the exchange
of vast amounts of information, goods, and services that can increase effi-
ciency and thus customer awareness and loyalty, but these advantages can
also turn into threats: Customers may be delighted to flip through dozens of
virtual sales racks and make immediate on-line purchases, but they certainly
do not want unauthorized parties to appropriate their credit card numbers
and other personal data, or interfere with their ability to access the Web site in
the first place. This problem is a very serious one for many companies. A re-
cent survey by CSI-FBI found that the Internet was the point of attack in 74
An earlier version of this paper was presented at WISE 2002 (Workshop on

Information Systems and Economics), Barcelona, and AAA-IS 2003 (American
Accounting Association–Information Systems), San Diego. The authors thank the
workshop participants as well as Hal Varian, Rajiv Banker, Ross Anderson, Radha
Mahapatra, Hasan Cavusoglu, and seminar participants at the University of Texas
at Dallas for encouraging them to pursue this study and for their comments on
earlier drafts of the paper.
International Journal of Electronic Commerce / Fall 2004, Vol. 9, No. 1, pp. 69–104.
Copyright © 2004 M.E. Sharpe, Inc. All rights reserved.
1086-4415/2004 $9.50 + 0.00.
70 CAVUSOGLU, MISHRA, AND RAGHUNATHAN
percent of hacking incidents in 2002, a rise from 38 percent in 1996 [61]. Ac-
cording to the Computer Emergency Response Team (CERT) Coordination
Center, the number of Internet attacks on business has almost doubled every
year since 1997 [16]. Not surprisingly, companies that do e-business are seek-
ing ways to exploit the interconnectivity of networks, encouraging and even
improving “open access,” but still maintain a security firewall.
Public awareness of security breaches increased dramatically when high-
profile Internet companies like Amazon, eBay, and Yahoo were hit by denial-
of-service (DOS) attacks in February 2000. Software developers are cognizant
of the need for secure products. In 2002, Microsoft took the unprecedented
step of ceasing development of new Windows operating system software for
an entire month and sending the company’s 7,000 systems programmers to a
special security training program [49]. In a memo addressing the incident, the
company’s president, Bill Gates, announced that security is now ”more im-
portant than any other part of our work. If we don’t do this, people simply
won’t be willing—or able—to take advantage of all the great work we do.
When we face a choice between adding features and resolving security issues,
we need to choose security. Our products should emphasize security right out
of the box” [15]. Gates’s memo did not estimate the cost of security breaches
to Microsoft, but it shows that a major software developer values features
related to security in its products.
Even as firms have come to recognize the importance of security, assessing
its economic value has proved to be challenging. Traditionally, organizations
have regarded security as a kind of insurance policy that mitigates conse-
quences rather than prevents them. According to Ron Knode, Computer Sci-
ences Corporation’s global director of managed security services, “While most
IS professionals recognize the benefits of protecting and securing data, the
business leadership in the organization still sees security as a ‘nice to have’
rather than ‘need to have.’ It is not until something goes wrong before percep-
tions change. The fact is, it costs far less to establish the right security mea-
sures at the outset than it does to recover from a breach in security” [25]. His
words proved true—three months later—when the February 2000 DOS attack
struck several firms.
The true cost of a security breach is manifold. Security lapses can lead to the
loss of consumer confidence and trust, over and above the lost business and
exposure to third-party liability. In a recent survey by Media Metrix, only 12.1
percent of U.S. companies with a Web presence cited direct financial loss as a
concern in security breaches, but more than 40 percent cited consumer trust
and confidence [58]. Although it may be impossible to directly quantify the
costs associated with the breaching of consumer trust and confidence, an indi-
rect estimate is possible on the basis of the capital market valuations of firms.
The present research has the following specific objectives:
1. To quantify the costs associated with IT security breaches using
market valuation data.
2. To explain the factors that affect the cross-sectional variations
related to breach costs.
3. To document the information-transfer effect of security breaches on
security developers’ market valuations.
INTERNATIONAL JOURNAL OF ELECTRONIC COMMERCE 71
The research in this paper is related to two previous studies of IT failures

and capital markets. Ettredge and Richardson investigated the stock market
reaction to the February 2000 DOS attacks and found that Internet firms suf-
fered market reactions more severely than did traditional firms [30]. Theirs
was the first study to measure the effects of security breaches on capital mar-
kets. Bharadwaj and Keil, who studied the impact of announcements of IT
failures, including DOS attacks on capital markets, found a significant drop in
the market value of firms that experienced such failures [10]. The present re-
search, like the work of the aforementioned authors, employs the event-study
method, but as far as is known, it is the first large-scale examination of the
effect of security breaches on capital markets. It differs from these two earlier
works in the following respects. First, unlike Ettredge and Richardson, it does
not restrict the analysis to DOS attacks, but considers security breaches of all
types. This makes it possible to investigate the differential effects of DOS as
compared with other types of attack. Although the February 2000 incident is a
landmark in the IT security domain because of the wide scope and publicity of
the attack, the paper analyzes the impact of security breaches in 1996–2001
and does not isolate the February 2000 incident. Next, whereas Bharadwaj
and Keil study security breaches among several other types of IT failure (they
also consider DOS attacks only), the present paper focuses exclusively on se-
curity breaches. For the purposes of the study, a security breach is defined as a
malicious attempt to interfere with a company’s business and its information.
Thus, the study does not consider accidental events and glitches that may af-
fect the firm. In what is, perhaps, its most important departure from the afore-
mentioned two studies, the present analysis is not limited to one segment of
the market but also looks at certain repercussions of an attack beyond the
breached companies. The effects of security breaches on breached firms are
analyzed and so are the information-transfer effects produced by the breaches
on security technology firms. The present paper is the first study in the infor-
mation systems (IS) literature to investigate information-transfer effects.1
The research found that announcements of Internet security breaches were
negatively associated with the market value of the announcing firms. The
breached firms, on average, lost 2.1 percent of their market value within two
days of the announcement. This translated into a $1.65 billion average loss in
market capitalization per breach based on the mean market value of the firms
in the data set. The research also found that
1. Breach cost is higher for “pure play,” or Internet-only, firms than for
conventional firms.
2. Breach cost increased during the study period.
3. Security breaches are costlier for smaller firms than larger firms.
4. Breach cost is not significantly different across breach types.
The effects of security breaches were not restricted to the breached firms,
however. The market value of security technology firms was positively asso-
ciated with the disclosure of a security breach. The security firms in the sample
realized, on average, an abnormal return of 1.36 percent within two days after
the announcement. This produced, on average, a total gain for security firms
of $1.06 billion in a two-day period.
Transitory Long-term
Tangible Low High

Intangible High Very high
Table 1. Degree of Uncertainty in Estimation of Costs.
Theory and Hypotheses Development
In order to derive hypotheses, the study identified the different types of costs
associated with security breaches and their relationship to firm value under
the efficient market hypothesis.
Cost of Security Breaches
The costs of security breaches can be broadly classified as transitory (or short-
term) costs, incurred only during the period in which the breach occurs, and
permanent (or long-term) costs, incurred over several periods.
The transitory costs of security breaches include lost business and decreased
productivity resulting from the unavailability of the breached resources; la-
bor and material costs required to detect, contain, repair, and reconstitute
breached resources; costs associated with evidence collection and prosecu-
tion of the attacker; costs related to providing information to customers and
the public; and other media-related costs [26].
Permanent, or long-term, costs have more far-reaching effects on the
breached firm’s future cash flow. These costs are related to the loss of custom-
ers who switch to competitors, inability to attract new customers due to per-
ceived poor security, loss of trust of customers and business partners, legal
liabilities arising from the breach, and the cost of attackers’ access to confi-
dential or proprietary information. Perceptions of increased business risk may
also translate into increased insurance costs for the firm and higher capital
costs in debt and equity markets.
The costs incurred as a result of breaches can be further classified as tan-
gible or intangible. It is possible to estimate the cost of lost sales, material and
labor, and insurance, but costs related to trust are difficult to calculate. None-
theless, these intangibles are extremely important in the measurement of the
overall cost of security for business. Table 1 blocks out the four types of costs
and the degree of uncertainty associated with dollar estimates of each type.
The magnitude of the four types of costs is likely to vary based on the breach
type and business type.
Investors’ Beliefs and Firm Values Under an Efficient Market

Hypothesis
The study constructed a model for valuating firms based on the efficient mar-
ket hypothesis and investors’ belief revision. Consider a firm whose value at a
time period t is denoted by Vt. Assuming that the firm will liquidate at the
terminal period T, the value of the firm can be expressed as the discounted
value of expected future cash flows at time t conditioned on all the informa-
tion available to the market until time t.2 Thus,
 
 G 
 T c i ηt 
Vt = Et ∑ i ,
 i=t ∏ (1+ rjt )  (1)
 
 j=t 
G
where ci ηt is the net cash flow in period i conditioned on all the information
available to the market until date t from the beginning of the firm’s existence.
G G
ηt = {ηt−1 , ηt }, with ηt the new information flow to the market during a time
from t – 1 to t. (Note that ηt itself can be multidimensional.) rjt represents the
term structure of the interest rate for the firm in a period j at some time t. Et
acts as the expectation operator at a time t (i.e., expectation with respect to
investor belief about cash flow distribution at time t).
The change in firm value between periods t and (t + 1) is
   
 G   G 
 T ci ηt+1  T c i ηt 
∆V = Et+1 ∑ i − Et ∑ i . (2)
 i=t+1 ∏ (1+ rjt+1 )   i=t ∏ (1+ rjt ) 
   
 j=t+1   j=t 
Given that the definition of period is arbitrary, if the length of the period is
shrunk such that ηt + 1 is simply the breach information during periods t and
t + 1, then ∆V represents the change in the firm’s value because of this firm-
specific event. Because the firm is part of a market, ∆V includes any change in
firm value that is related to market forces. Let ∆Vb be the change in value
produced by a firm-specific security breach event, and let ∆Vm be the change
in firm value because of a concomitant market event. Thus, for a breached
firm,
∆Vb = ∆V −∆Vm . (3)
Several interesting observations are obtained from this model. First, both
the change in the breached firm’s value and the change in the market value
are observable during the event period. The change in firm value is derived
due to breach from these two observed values. Second, the change in the
breached firm’s value subsumes all types of breach costs mentioned above,
and the different types of costs cannot be distinguished. However, an indirect
approach can be used to get an estimate of these costs: The types of costs
likely to be associated with the breach event are specified ex ante for the firm
under consideration using specific information about the type of breach and
the attributes of the firm and rational arguments. For example, it is reasonable
to assume that most of the costs associated with a DOS attack will be tangible
short-term costs related to loss of business due to unavailability of the breached
information resources. However, in most cases, it may be nearly impossible to
separate the different types of breach costs. Third, in view of the fact that
investors change their beliefs based on all the information that is known to
date, the breach value will be correlated with the known characteristics of the
firm and breach type and any other information available to the investors.
Thus, in a cross-sectional analysis, the calculated breach cost can be regressed
against these attributes to better determine their effects on the breach cost. In
addition, it is important to note that because investors revise their beliefs based
on their most current information, the same information event can generate
different belief revisions for investors at different times.
Development of Hypotheses
In keeping with the discussion of breach costs and firm value under the effi-
cient market hypothesis, several testable propositions are derived in the dis-
cussion that follows.
Impact of Security Breaches on Market Value of Breached Firm
As discussed above, a security breach is multifaceted and can have both tan-
gible and intangible costs. Whereas most tangible costs are immediate or short-
term, the intangible costs can have a long-term effect on the firm’s expected
future cash flows. Anecdotal evidence suggests that the impact of a security
breach on the breached firm’s value can be significant. For example, immedi-
ately following the February 2000 DOS attack, Yahoo, eBay, and Buy.com lost
15 percent, 24 percent, and 44 percent, respectively, of their market value [4].
The relationship between IT security and market valuations of firms doing
business on the Internet can be traced to the trust of customers who do busi-
ness with the firm through the Internet. Customer trust assumes more signifi-
cance in e-business because of concerns related to data privacy. A customer
may be unwilling to transact business with sites perceived to be insecure. A
security breach can irrevocably damage the trust and confidence necessary to
build a long-term relationship with the customer. In the Internet era, charac-
terized by much competition and little loyalty, dissatisfied customers can switch
to competitors that are just a click away. Thus, perceptions of lax security can
have a profound financial impact on a firm. Security problems may signal to
the market that the firm is not concerned about customer privacy or that its
internal security practices are poor, and this may lead investors to question
the firm’s long-term performance [32].
As described in the model given above, investors revise their expectations
based on new information in announcements. Investor expectations are re-
flected in the value of the firm. If investors view a security breach negatively,
believing that the transitory and long-term costs resulting from it will sub-
stantially reduce expected future cash flows [31], then one may expect a nega-
tive abnormal stock market return near the day of the announcement.
H1: A firm’s announcement of an Internet security breach is negatively
associated with its abnormal stock return.
Determinants of Cross-Sectional Variance
The discussion in this section develops hypotheses to explain the cross-sec-

tional variation in abnormal returns caused by security breaches. The deter-
minants of cross-sectional variation in abnormal returns are classified into
three categories: firm-specific factors, context-specific factors, and event-spe-

cific factors. Firm-specific factors, such as firm type and firm size, affect the
magnitude of reaction in capital markets. Time (i.e., date security breach oc-
curred), defined here as a context-specific factor, is used as a variable to con-
trol for changes in investor beliefs correlated with the passage of time. Breach
type is defined as an event-specific factor. Because investors presumably know
these factors when they revise their expectations of future cash flows, these
variables will correlate to the cross-sectional variation in the change in firm
value triggered by a breach event.
1. Firm-specific factors comprise the first group of determinants to be
discussed.
Firm type. Firms doing business on the Internet are typically grouped into
two categories: conventional firms and net firms. Conventional firms, some-
times referred to as brick-and-mortar or click-and-mortar firms, have a cus-
tomer base in traditional markets and thus do not rely solely on the Internet to
conduct business. Brick-and-mortar firms (e.g., Coca-Cola) do not use the
Internet as an additional channel to sell their products, but they use it as a
marketing tool to communicate with their customers. Click-and-mortar firms
(e.g., Borders) use the Internet as another channel—in addition to their physi-
cal stores—to sell their products. Internet firms, also called pure play or
Internet-only companies, include firms like Amazon and eBay, which rely ex-
clusively on the Internet to sell their products and services.
Although security breaches can impose short- and long-term costs on both
types of firms, the intangible long-term costs are more severe for Internet firms.
These firms rely solely on the Internet for their survival, so information secu-
rity is not an attractive optional feature but an essential ingredient for success.
Outages caused by DOS attacks mean lost revenues and lost opportunities,
because customers cannot make intended purchases. Attacks that penetrate
the confidentiality and integrity of customer information may lead to liability
suits and long-term competitive disadvantage. All of these costs are nearly
impossible to estimate ex ante.
Consumer concern about on-line security has been confirmed in several
surveys. Respondents in a Rockbridge Associates study expressed suspicions
about the security of on-line transactions: 58 percent did not consider any on-
line transaction safe, 67 percent were not confident about conducting busi-
ness with a company that could only be reached on-line, and 77 percent thought
it was unsafe to provide a credit card number over a computer [7]. A study by
the Angus Reid Group revealed that many Internet users have never shopped
on-line, because they fear their credit card information will be leaked or sto-
len [2]. Consumer fears about security and security breaches impose intan-
gible costs on firms, because they contribute to the loss of consumer confidence.
Conventional firms are relatively less affected than net firms by security
breaches on the Internet. Firms that use the Internet simply to provide infor-
mation to the public will incur little damage from DOS attacks on their Web
sites, and even those that conduct business over the Internet will not be com-
pletely shut down by such attacks. If, for example, a security problem shuts
down the Borders Web site, the company can continue to sell books through
its physical stores. This is not true for Amazon. If its Web site is down, there is
no other way for customers to transact business with the company. Besides,
the intangible costs in terms of customers’ and partners’ loss of trust in the
organization would probably be much more severe for net firms. Thus, even
if the dollar cost is the same in terms of time and other resources spent on
fixing the breach and getting the system up and running, intangible long-
term costs are likely to be far greater for Internet firms than for non-net firms.
Based on the above arguments, one may expect the impact of security breaches
to be higher for net firms than for conventional firms. Because the change in a
firm’s value due to a breach captures both the short- and long-term tangible
and intangible costs, one may posit a significant long-term intangible cost to
net firms, a cost that will be reflected in stock price returns. In other words,
cumulative negative abnormal returns are higher for Internet firms than for
non-net firms.
H2: The magnitude of abnormal negative returns for Internet security
breaches is larger for net firms than for conventional firms, ceteris paribus.
Firm size. It is well documented in the finance literature that the expected
return for small firms exceeds the expected return for large firms even after
accounting for beta, or market, risk [6, 33, 63]. Fama and French attribute this
discrepancy to possible distress risks in small firms [33]. With greater access
to capital markets, lower costs for capital, multiple sources of income, diversi-
fied market products, and brand-name recognition, large firms can absorb
negative economic and financial shocks more easily than small firms. Secu-
rity breaches can be viewed as negative economic shocks. Large firms are
likely to handle these shocks better than small firms.
In the strategic management literature, the resource-based theory of firms
is widely used to explain competitive advantage. This definition of a firm as a
broader set of resources that can be used to accomplish organizational goals
can be traced to the seminal work of Penrose [59].3 Resources are broadly de-
fined as tangible or intangible assets that are tied semipermanently to the firm
and may include brand names, skilled personnel, technological skills, pat-
ents, machinery, capital, and efficient procedures [14]. Large firms are likely
to have more of these resources than small firms: more capital to work with,
more slack resources to deploy in case of a security breach, such as backup
Web servers and IT staff, perhaps even more highly skilled IT personnel than
small firms. These differences can produce asymmetries in the impact of secu-
rity breaches. Both the distress risk issues cited by Fama and French and the
resource-based theory indicate that large firms may be better able to with-
stand the negative impacts of security breaches than small firms.
In addition, empirical studies in the IS domain have documented similar
relationships between firm size and abnormal returns. Bharadwaj and Keil
found an inverse relationship between firm size and (negative) abnormal stock
returns when they examined the effects of IT failures [10]. Hendricks and
Singhal reported similar results for supply-chain problems [39]. In a study of
the impact of IT investments, Im and colleagues observed that (positive) ab-
normal returns were negatively related to firm size [40].
H3: An abnormal (negative) stock market return due to an Internet
security breach is larger for smaller firms than for larger firms, ceteris
paribus.
2. The discussion of determinants now proceeds to event-specific factors.
Nature of attack (breach type). IT security breaches are associated with three
primary categories of attack: access attacks, modification attacks, and DOS
attacks [51]. Each type of attack compromises one or more of a company’s
security objectives: confidentiality of customer information, integrity of infor-
mation, and availability of applications and services.
In an access attack, the attacker attempts to gain unauthorized access to
information. Access attacks may occur through technical means that exploit
a vulnerability in the system, so that the attacker gains access to customer
information purely electronically. They may also be made through “social
engineering”—ploys or scams by which the attacker attempts to extract un-
authorized information from the customer (e.g., by posing as a customer rep-
resentative on the telephone and requesting credit card or other information).4
A modification attack is one in which the attacker attempts to make illegal
insertions or deletions of information. This type of attack compromises the
integrity of the information.
DOS attacks deny the use of resources, applications, or information to le-
gitimate users of the system.5 They are arguably the most critical security is-
sue on the Internet, particularly for e-commerce services.6 If the site is not
available, revenue is lost, resulting in a direct hit to the bottom line. Customer
dissatisfaction can grow quickly.
On the upside, DOS attacks have relatively brief durations and do not de-
stroy data. Although obviously disruptive, attacks on availability only affect
site accessibility, making them less damaging than other forms of attack that
destroy, manipulate, or expose programs and sensitive information on a wide
scale. DOS attacks do not expose firms to the potentially large third-party
liability associated with breaches of privacy and unauthorized leaks of cus-
tomer information. Thus, although tangible costs resulting from lost business
can be substantial after a DOS attack, the intangible costs are likely to be higher
with other types of attack. Because the overall cost of a breach includes both
types of cost, it is difficult to determine whether capital markets consider avail-
ability breaches more serious than other breach types. Thus, a directional re-
sult is not proposed in the next hypothesis.
H4: The magnitude of abnormal negative returns that result from Internet
security breaches will differ between availability and all other types of
attack, ceteris paribus.
3. Context-specific factors are the final body of determinants to be discussed.
Time. In the early days of the Internet, few security-related incidents oc-
curred, and firms and investors paid little heed to the possibilities of access,
integrity, or availability attacks. However, where once security was consid-
ered an unnecessary burden, in recent years, these issues have become a pri-
mary concern, as more firms experience security breaches. The pattern of
increase can easily be seen in the number of security incidents reported to
CERT. In 1996, there were only 2,573 reports of security-related incidents. This
number jumped to 21,756 in 2000, and to 52,658 in 2001 [16]. The popular
press has begun to publish more news about Internet security. A key word
search of the term “Internet security” in Lexis/Nexis revealed 97 articles in
1996, 696 articles in 2000, and 625 in 2001. Recent attacks on high-profile Web
sites like Amazon, eBay, Yahoo, and Etrade have contributed to heightened
awareness of Internet security.
Investors cannot have perfect insight into the exact costs of a particular
security breach, but as more is discovered about the cost of security breaches
in general, investors in due course may revise their beliefs about the costs.
The eventual change in reaction to similar events has been documented by
several IS researchers. Bharadwaj and Keil find that (negative) abnormal re-
turns due to IT-related project failures are positively associated with the pas-
sage of time [10]. Similarly, Subramani and Walden show that market reactions
to e-commerce announcements changed from positive to negative over time
[69]. In light of this, the research design controls for time. If a learning curve is
involved in investors’ gaining understanding of the implications of security
breaches, one would expect the response to security breaches to change over a
time interval [1].7 Time is used as a control variable to proxy for such belief
revision over time.8 Earlier researchers, such as Mikhail, Walther, and Willis
[52], used this variable to control for similar effects.
Relationship Between Internet Security Breaches and Market Value of

Internet Security Firms (Information Transfer)
Announcements of security breaches convey information to investors about

Internet security developers. Anecdotal evidence suggests that there is a link
between security breaches and the market value of Internet security firms.
Following the news of the February 2000 DOS attacks, five different Internet
security stocks climbed more than 20 percent, and one firm, WatchGuard Tech-
nologies, gained 46 percent. When Microsoft was hacked almost a year later,
the stock prices of security firms also increased [55]. A series of attacks against
Web firms in the last few years proves that this is not a temporary issue. As
firms invest more in security, demand for security products increases. An es-
timate by IDC predicts that the worldwide market for IT security products
will reach $21 billion in 2005, from $6.7 billion in 2001. Gartner predicts that
investments by U.S. companies in information security will increase from the
current 0.4 percent of revenue to 4 percent of revenue by 2011, a 1,000 percent

increase [64].
Information transfers are said to occur if announcements made by one group
of firms contemporaneously affect the returns of another group of non-an-
nouncing firms [66]. Previous studies have documented information trans-
fers in various settings, such as earnings announcements, sales announcements,
and management forecasts [3, 5, 20, 34, 35, 37, 56]. Most information-transfer
studies concentrate on intra-industry information transfer, but Olsen and
Dietrich, and Chang, Mishra, and Huang consider interindustry information
transfers [17, 56]. Olsen and Dietrich show that monthly sales announcements
by retailers lead to statistically significant changes in the stock prices of retail-
ers and their suppliers. Chang, Mishra, and Huang studied information transfer
across industries in the supply chain and the characteristics of the supply chain,
such as relative dependence, level of vertical integration, and type of news
(i.e., good vs. bad) that affects the magnitude of information transfer.
The notion of information transfer in the present paper is similar to the one
used by Olsen and Dietrich and by Chang, Mishra, and Huang in that it posits
information transfer between firms that suffer from breaches and firms that
are potential suppliers of security products to breached firms [17, 56]. Follow-
ing both empirical and anecdotal evidence, it is expected that announcements
of security breaches will have a positive impact on the valuation of the stocks
of security firms.
H5: Announcements of Internet security breaches are positively associated
with abnormal stock market returns of Internet security firms.
Data Set and Methodology
Sample Selection
For the purpose of analyzing how announcements of security breaches affect

capital markets, “event” is defined in this study as the first public disclosure
to the media of a security breach of a firm. The study covered security breaches
that occurred between January 1, 1996, and December 31, 2001. The samples
of announcements of security breaches came from three different news sources:
Lexis/Nexis, and the technology portals CNET and ZDNET. Lexis/Nexis was
chosen because the databases it covers includes the major U.S. newspapers.
CNET and ZDNET were chosen as alternative databases to supplement the
articles found in Lexis/Nexis, because they are highly regarded global sources
of information for the technology industry. After careful examination of sev-
eral articles about Internet security breaches, the on-line search features of
these sources were used to search for announcements using the key words
“attack,” “breach,” and “break-in” in the same search string as the words
“hacker,” “Internet,” and “security.” This search resulted in 2,563 articles for
potential events.
Every news article that mentioned specific breaches was considered in the
study. A news article might refer to a press release by a breached firm or might
include news about an apparent security breach, such as a DOS attack or a
Web defacement (see Appendix A for an illustrative sample). Consistent with

the literature, an announcement that contained news about security breaches
at multiple firms was counted as announcing multiple events, each one relat-
ing to one of the firms involved [70]. The data search included duplicate items,
because data were gathered from different sources. When there was more than
one announcement about the same security breach, the earliest announcement
was retained, and the others were eliminated. Also eliminated were announce-
ments related to IT project failures, such as implementation and operation
system failures resulting from sudden malfunctions in software or hardware,
because these were unrelated to security. At the end of this step, 225 events
corresponding to security breaches were identified from the 2,563 articles.
The next step eliminated announcements of attacks against firms that were
not publicly traded—government agencies, nonprofit organizations, and pri-

vately held companies. Also removed from the data set were publicly held
companies not traded in the United States. These steps reduced the number of
events in the database to 78.
After these screenings, the remaining data set only included announcements
pertaining to firms traded on U.S. capital markets. These firms were then
matched with data available on the University of Chicago’s Center for Re-
search in Security Prices (CRSP) daily common-stocks returns tapes. Firms
that did not have return data in CRSP, such as those traded in the OTCBB
market [23], were matched with return data in either the NASDAQ Quote or
Yahoo Quote archives. A few announcements were eliminated from the data
set because sufficient historical return data necessary for a meaningful analy-
sis were unavailable. The remaining announcements were checked against
confounding factors, such as dividends, mergers and acquisitions, earnings,
or other significant public announcements, that could undermine the results
of the study. Consistent with the literature, the check for confounding factors
was conducted on the day before, the day of, and the day after the security
breach announcement. After this last check, the final database consisted of 66
security breach announcements in the period 1996–2001.
Sample Coding
Each announcement and all related firm characteristics were carefully exam-
ined to operationalize a classification scheme to test hypotheses 2 through 5.
Following the convention in other event studies, market value was used as a
surrogate for firm size [10, 40]. The natural logarithm of market value (in mil-
lion dollars) of the firm at the end of the year immediately preceding the event
date was used as a measure of firm size. Market value data were collected
from the Compustat database.
Internet.com’s Internet Stock ListTM and Morgan Stanley Dean Witter’s
Internet Companies List were used to code the firms in each announcement
as either conventional or net firms. These two lists are the most comprehen-
sive lists of Internet firms. They only include companies that are solely in
Internet-related business. The same lists were used in prior studies [30, 38,
71]. In the present study, a firm was classified as a net firm if it was listed on
both lists. If it did not appear in either list, it was coded as a conventional firm.
There were two firms, listed in one list but not the other. They were classified
as conventional firms because a detailed examination revealed that they gen-
erated a significant portion of their revenues from non-Internet channels. Of
the 66 events, 31 were coded as announcements by Internet firms, and the
remaining 35 as announcements by conventional firms.
To code the nature of the breach as either an availability breach or other
breach, each announcement was carefully examined to determine whether
the security breach resulted in the loss of availability of a service, application,
or information. Two of the authors independently categorized the security
breaches. The coders were in agreement in 96.96 percent of the cases. This
level of agreement indicates a very high level of intercoder reliability. Incon-
sistencies in categorization were resolved through discussions. Of the 66 events,

34 were coded as announcements for availability attacks, and the remaining
32 as announcements for other attacks.
Sample Description
Table 2 presents the distribution of security breach announcements in the fi-

nal data set. The frequency of security breaches ultimately increased. There
were only two announcements in 1996. The number jumped to 21 in 2000 and
26 in 2001. About 29 percent of the 66 announcements were made between
1996 and 1999, and the remaining 71 percent between 2000 and 2001. The
increase in the number of announcements may be attributed to several rea-
sons, including an increase in the number of firms with a presence on the
Internet, greater awareness of security, and an increase in the media coverage
of IT security.
Table 3 shows descriptive statistics for firms in the sample of 66 events
based on data from the most recent year before the announcement date of
security breach. The mean asset value was $38.5 billion, and the mean market
value was $78.3 billion. Higher mean values compared to median values indi-
cates that the sample is skewed toward larger firms.
Selection of Internet Security Product Firms
The list of Internet security firms was compiled from two sources:
INFOSYSSEC, the security portal for information system security, and Infor-
mation Security Magazine, the leading magazine for the security industry.
INFOSYSSEC provides a section about top security companies and their stock
quotes that lists more than 40 publicly traded security firms [42]. This list was
augmented with the list of security firms nominated for the 2002 Information
Security Excellence Awards by Information Security Magazine [41]. These awards
are given to companies that offer products and services recognized as leaders
in the information security field.
A total of 128 security firms was identified after the firms in these two
sources were combined. When the duplicates were removed, the list contained
Number of
Year security breaches
1996 2
1997 5
1998 6
1999 6
2000 21
2001 26
Table 2. Security Breach Announcements, 1996–2001.

99 unique firms. The list was screened to identify the private firms and for-
eign security firms not traded in the United States—this process eliminated
43 firms from the list. After a careful investigation, any of the remaining firms
whit core business that was not Internet security were also removed. The busi-
ness description of each firm provided in Yahoo Finance was used for this
purpose. If security was not mentioned in the description, the firm was dropped
from the list. This process removed 16 firms from the list. The final list after all
these checks comprised 40 security firms traded in U.S. capital markets (see
Appendix B for details).
In the next step, a security firm sample was compiled for each event—that
is, for each announcement of a security breach. We included a security firm in
an event’s security firms sample only if it had been a publicly traded com-
pany for at least 160 days before the day of the event. The exact composition
of events’ security firm samples varied with time, because the number of se-
curity firms that qualified for inclusion (out of 40 firms) in the security firm
sample varied with time. The mean number of security firms for each event
was 31.12, the minimum was 17, the median was 36, and the maximum was
38. As in the case of the breached firms, these firms were then matched with
their return data in CRSP.
Statistical Methodology
The event-study methodology was used to assess the impact of announce-

ments of security breaches on capital markets. An event study seeks to deter-
mine the effect of an announcement (the event) on the stock prices of firms.
This method has been employed extensively in the accounting and finance
literature to study the effects of an assortment of events, ranging from corpo-
rate acquisitions to joint venture formation to CEO successions [28, 36, 43, 47,
67, 68]. The first application of this methodology in the IS literature was the
study by Dos Santos, Peffers, and Mauer, who examined the effect of IT in-
vestment announcements on market values of firms [29]. More recently, the
effects of various types of IS-related announcements on capital markets have
been examined, including e-commerce initiatives, IT failures, IT investments,
dot-com name changes, and newly created CIO positions [10, 18, 23, 40, 70].
Standard
Mean Median deviation Minimum Maximum
Total assets (in million $) 38,558.9 5,348.0 109,158.2 37.8 642,191.0

Sales (in million $) 12,199.6 4,357.2 16,282.5 11.0 64,826.0
Net income (in million $) 1,655.1 344.7 3,134.2 –1,457.6 11,797.0
Market value (in million $) 78,327.0 23,333.7 113,928.2 158.7 460,770.5
Number of employees (in thousands) 38.5 12.7 54.8 0.1 260.0
Table 3. Descriptive Statistics for Breached Firms.

INTERNATIONAL JOURNAL OF ELECTRONIC COMMERCE
83
The event of interest in the present study is the announcement of an Internet

security breach in a firm. To determine whether an announcement affects a
firm’s stock price, for each firm, the study first estimated what the return of
the stock would have been had the event not occurred, that is, the normal
return. Consistent with other event studies, the market model used in the
study was one that predicts a linear relationship between the market return
and the return of a stock.9 The model is specified as
Ri ,t = αi + β i Rm ,t + εi ,t , (4)
where Ri,t is the return of stock i on day t; Rm,t represents the return on the
market portfolio on day t; αi and βi stand for the intercept and slope param-
eters, respectively, for firm i; and εi, t symbolizes a disturbance term for stock i
on day t, with the usual ordinary least squares (OLS) properties.
The NASDAQ composite index was chosen as the market index. It includes
more than 4,000 companies, more than most other stock market indices. Be-
cause it is broad-based, the composite is one of the most widely followed and
quoted market indices. Most of the firms in the sample were technology firms,
so the use of the NASDAQ composite index (characterized as a technology
index) as the market index was appropriate. Previous event studies on Internet
firms also used the NASDAQ as the market index (e.g., [62]).10
The market model was used to estimate the intercept and slope parameters
for each firm in the sample. The sizes selected for the estimation window and
the event window were based on previous event studies. The estimation pe-
riod ranges typically from 120 days to 200 days [10, 29, 40, 70]. The study used
an estimation window of 160 days. The estimation window started 160 days
before the announcement day and ended one day before the announcement
day. In event studies, the selection of the event window over which the effects
of an announcement are examined is crucial [50].11 A one-day event window
is usually preferred. In practice, the event window is often expanded to two
days, the day of announcement (t = 0) and the day after the announcement (t
= 1) [13, p. 151]. This is done to capture the price effects of announcements
that occur after the stock markets close on the announcement day. If the an-
nouncement is made before the markets close, any response to new informa-
tion in it will be reflected on the announcement day. However, if the
announcement is made after the markets close, the effects will be reflected in
the stock price on the day following the announcement. Some studies included
the period before the announcement day in the event window in order to
capture the possibility of information leakage to markets before the announce-
ment day [70]. Because security breaches are generally unanticipated, how-
ever, the event window in the present study did not incorporate any period
before the announcement. Consistent with the above arguments, a two-day
event window was used in the analysis. To the extent that the actual announce-
ment day preceded the day used in the sample, the tests were biased toward
accepting the null hypothesis of no abnormal returns in the event window.
The coefficient estimates, α{i, β{i, from regression of Equation 4 were used to
predict the expected return over the event window. Then the abnormal return
for firm i on day t of the event window was calculated, following McWilliams
and Siegel [50]
( )
ˆ i + βˆ i Rm,t .
ARi ,t = Ri ,t − α (5)
Abnormal returns (also called excess returns) are deviations of realized re-
turns from normal returns. They are unbiased estimates (in the return form) of
changes in the market value of the firm during the event period that are attrib-
uted to investors’ reactions to information contained in an announcement [29].
The standard errors are calculated as in Subramani and Walden [70]:
  
 
(Rm ,t − Rm ) 
2
 1
var ( ARi ,t ) = si21+ + −1 
 160 2 (6)

  ∑ ( Rm ,t − Rm ) 

  t=−160 
where si2 is the residual return variance from the estimation of the market
model on the interval of 160 days before the event window, R|m is the mean
market return on the market index over the estimation window, and Rm,t is the
return on the market index on day t in the estimation window.
Assuming that abnormal returns are independent of time, for firm i, the
cumulative abnormal return (CAR) and the variance of the cumulative abnor-
mal return are the sum of the individual abnormal returns and variances over
the event window, respectively. Thus,
1
CARi = ∑ ARit (7)
t=0
and
1
var (CARi ) = ∑ var ( ARit ). (8)
t=0
These are aggregated across all events to draw an overall inference as
1 N
CAR = ∑ CARi (9)
N i=1
along with
N
1
var (CAR ) = 2 ∑ var (CARi ). (10)
N i=1
Cumulative
Abnormal returns abnormal returns
t=0 t=1 (t = 0, t = 1)
Mean –0.0086 –0.0123 –0.0209

Minimum –0.1806 –0.1803 –0.3609
First quartile –0.0225 –0.0207 –0.0294
Median –0.0025 –0.0050 –0.0144
Third quartile 0.0141 0.0067 0.0185
Maximum 0.0655 0.0394 0.0892
Table 4. Descriptive Statistics of Abnormal Returns and Cumulative

Abnormal Returns (n = 66).
A student’s t-test was used to test the alternative hypothesis that the mean
CAR over the event period significantly differed from zero:
CAR
t= t( a, df =N−1) .
var (CAR ) (11)
Results
Effect of Security Breach Announcements on Announcing

Firms
Table 4 presents the abnormal returns for each day in the event window and
the cumulative abnormal returns over the event window resulting from the
announcements of Internet security breaches. The median and mean cumula-
tive abnormal return over the event period and the median and mean abnor-
mal return for each day in the event window were negative. On announcement
day (t = 0), an average of –0.86 percent abnormal returns was observed. The
stocks realized, on average, an abnormal return of –1.23 percent on the day
following the announcement (t = 1). This gave rise to a –2.09 percent cumula-
tive abnormal return over the event window. Compromised firms lost, on
average, around 2.1 percent of their market value within the two days sur-
rounding the events.12 This translated into a $1.65 billion average loss in mar-
ket capitalization per incident, based on the mean market value of the firms in
the data set.13
Table 5 presents the results to test H1. The mean abnormal returns are nega-
tive and statistically significant for each day in the event period. Over the
event window, the t-statistic for the mean cumulative abnormal return is al-
most 3, with a p-value of 0.00192, indicating that the mean cumulative abnor-
mal return is significantly different from zero. Thus, the null hypothesis of a
zero-mean cumulative abnormal return is rejected in favor of the alternative
hypothesis 1.
Frequency of
Event negative
window Mean t-value p-valuea returnsc z-valueb p-valuea
0 –0.008638 –1.74985 0.04243 36 –0.948630 0.171404

1 –0.012282 –2.49133 0.0076442 39 –1.894067 0.029108
0,1 –0.02092 –2.99862 0.00192 41 –1.823797 0.034091
Table 5. Event Study Results for Breached Firms.

a
p-values of one-tailed significance.
b
If T+ is the sum of the ranks assigned to positive CARs, and N is the sample size, then the test statistic is given by ( T+ – a)/b, which is distributed as
N(0, 1) for large samples, where a = N*(N + 1)/4, and b = N*(N + 1)*(2N + 1)/24.
c
Of 66 observations.
87
The results given above are based on the assumption that abnormal re-
turns and cumulative abnormal returns are normally distributed random vari-
ables. The Wilcoxon signed-rank test—a nonparametric test—was employed
to check the robustness of the results [53, p. 534]. As McWilliams and Siegel
pointed out, nonparametric tests are important to control for the effects of
outliers on the significance of results, because most event-study parametric
test statistics are sensitive to outliers [50]. The results in Table 5 indicate that
the signed-rank test statistic is negative (z = –1.8238) and significant (p = 0.0341),
confirming the results of the parametric test.
Determinants of Cross-Sectional Variance in Cumulative

Abnormal Returns
A multiple linear regression model, given in Equation 12, was set up to study
the relationship between cumulative abnormal returns and characteristics of
events. The model regressed the cumulative abnormal returns on hypothesized
variables and control variables, namely, firm type, firm size, nature of attack,
and time. This made it possible to test hypotheses 2 through 4.
CARi = β 0 + β1 ( FirmType )i + β 2 ( FirmSize )i + β 3 ( AttackType )i

(12)
+ β 4 (Time )i +εi .
The independent variables were operationalized as follows. Using the cod-

ing scheme discussed above, firm type was labeled using an indicator vari-
able. A value of 1 was assigned if the firm type was Internet , and 0 otherwise.
The size of the firm was coded as the log of the market value of the firm (in
millions) at the end of the year immediately before the event. The nature of
the attack was labeled 1 if the security breach was an availability attack, and 0
otherwise. Finally, time was captured using the number of years from the first
announcement date. That is, the date on which the first breach announcement
was made in the data set was chosen as the reference point and coded as zero,
and the time for each successive announcement was captured as the differ-
ence between the date of the successive event and the date of the first event in
units of year.14 Descriptive statistics and a correlation matrix for independent
variables are shown in Table 6.
Table 7 presents the results of the regression analysis. The overall model is
significant (F = 4.63; p = 0.0025) with an R2 of 0.2311 and an adjusted R2 of
0.1828. H2 is supported by the results reported in Table 7. The coefficient for
firm type is negative and significant (t = –1.71; p = 0.0461), indicating that
abnormal (negative) stock market returns due to Internet security breaches
are larger for net firms than for conventional firms. The results show, other
things being equal, that compared to conventional firms, the stocks of net
firms, on average, experience an additional 2.83 percent negative abnormal
returns.
In contrast to expectations, no support was found for H4. The supposition
that availability attacks have a different impact from other types of attacks was
Standard
Variable Mean deviation Minimum Maximum (1) (2) (3) (4)
(1) Firm type 0.47 N/A 0 1 1.00

(2) Firm size 9.88 2.09 5.07 13.04 –0.25 1.00
(3) Nature of attack 0.52 N/A 0 1 –0.18 0.21 1.00
(4) Time 3.30 1.34 0 5.03 –0.20 0.32 –0.06 1.00
Table 6. Descriptive Statistics and Pearson Product–Moment Correlations.

89
Predicted Parameter Standard

sign estimate error t pa
Intercept –/+ –0.1141 0.0433 –2.64 0.0106

Firm type (H2) – –0.0283 0.0166 –1.71 0.0461
Firm size (H3) + 0.0150 0.0042 3.60 0.0003
Nature of attack (H4) –/+ –0.0122 0.0164 –0.74 0.4597
Time – –0.0107 0.0064 –1.68 0.0488
Model R2 0.2331
Adj. R2 0.1828
F-value 4.63 (0.0025)b
Table 7. Results of Cross-Sectional Regression to Predict Variability in

Cumulative Abnormal Returns.
p represents one- (two-)tailed significance when sign is (is not) hypothesized.

a
Number in parentheses specifies the significance of F.

b
not supported by the data. The parameter estimate was negative but not sig-
nificant (t = –0.74; p = 0.4597). This implies that markets do not distinguish
between different types of security breaches. In other words, market partici-
pants apply a similar negative premium to all attacks regardless of type. H3,
which relates firm size to abnormal return, was strongly supported by the
data. The parameter estimate for the firm size was positive and highly signifi-
cant (t = 3.60; p = 0.0003), as expected. The result confirmed that smaller firms
lose more than larger firms in case of a security breach. The estimate of time
coefficient was negative and significant (t = –1.68; p = 0.0488), indicating that
investors reacted more harshly to the recent attacks.
Effect of Security Breach Announcements on

Internet Security Firms
The study assumed that the abnormal returns were uncorrelated in the cross-
section in order to analyze the effect of security breaches on announcing firms.
This was a reasonable assumption, because most of the announcements oc-
curred at different time intervals. The assumption made it possible to calcu-
late the variance of aggregated sample cumulative abnormal returns as the
sum of the variances of individual sample cumulative abnormal returns. The
assumption is not defensible, however, when the event windows overlap.
Because all the security firms experienced the same signal for possible infor-
mation transfer (i.e., an announcement of a security breach) on the same day,
the event windows overlapped perfectly. In this case, the assumption of no
covariance between the residuals (abnormal returns) of the security firms
would have been problematic [9]. Therefore, in this instance, it was undesir-
able to aggregate the estimated variances of the cumulative abnormal returns
of individual Internet security firms for a given security breach announce-
ment to estimate the variance of the mean cumulative abnormal return for
that event. Ignoring the contemporaneous correlation of residuals is even more
dangerous if the sample firms come from the same industry, as in this case,
because of the presence of positive industry cross-sectional return covariation

[57]. Hence, it is crucial to consider cross-sectional dependencies to avoid in-
correct inferences.
It was decided, because of these concerns, not to estimate the variance of
mean cumulative abnormal return from the estimated variances of sample
cumulative abnormal returns for each information-transfer event. Instead, only
the mean cumulative abnormal return was calculated for each event, CAR i .
Assuming that all CAR i s were random samples from an infinite population
with the same mean and variance, the following procedures were used to
estimate mean and variance:15
1 N
CAR = ∑ CARi (13)
N i=1
( ) ( )
N
1 2
var CAR = ∑ CARi − CAR , (14)
N * ( N − 1) i=1
where CAR is the mean cumulative abnormal return of security firms for
security breach i, and CAR is the average of all CAR i s. Therefore, the test
statistic is
CAR
t= t( α ,df =N−1) .
(
var CAR ) (15)
Table 8 presents the abnormal returns for each day in the event window
and the cumulative abnormal returns over the event window resulting from
the announcements of Internet security breaches. Both the median and the
mean of the cumulative abnormal returns over the interval and the abnormal
returns for each day in the event window are positive. An average of 0.71
percent abnormal stock return was observed on the announcement day (t =
0). The stocks realized, on average, an abnormal return of 0.65 percent on the
day after the announcement (t = 1). This gave rise to a 1.1356 percent cumula-
tive abnormal return over the event window. In other words, each security
developer realized, on average, a 1.36 percent return above the return expected
by a market model. Thus, in the two-day period, a total average gain of $1.06
billion could be attributed to the information-transfer effects of security breach
announcements on the market values of security developers.
The information-transfer effects of security breach announcements are re-
ported in Table 9. The positive abnormal return is significant (t = 2.712). The
nonparametric test is also significant (z = 2.028). These results support the
hypothesis that security breach announcements affect the values of announc-
ing firms and also of Internet security developers. The market value of secu-
rity firms increased in anticipation of the expected future gains from such
incidents.
Cumulative abnormal
Abnormal returns returns
Statistics t=0 t=1 (t = 0, t = 1)
Mean 0.0071 0.0065 0.0136

Minimum –0.0268 –0.0443 –0.0598
First quartile –0.0084 –0.0085 –0.0113
Median 0.0018 0.0030 0.0041
Third quartile 0.0168 0.0160 0.0302
Maximum 0.0789 0.1226 0.1349
Table 8. Descriptive Statistics of Abnormal Returns and Cumulative

Abnormal Returns (n = 66).
Sensitivity Analysis of Results
The robustness of the results to influential observations and model assump-

tions was evaluated by a series of sensitivity analyses. The first analysis dealt
with the effect of outliers on the t-tests (H1, H5). Following convention, data
points (events) that were outside the range of the mean, plus and minus three
standard deviations, were removed [19]. For breached firms, this resulted in
the elimination of one event from the data set. The cumulative abnormal re-
turn after this event was deleted turned out to be –1.57 percent and was sig-
nificantly greater than zero (t = –2.29; p = 0.012). For security developers, no
points were outside the limits. The nonparametric test controls for outliers, so
there was good reason for confidence that the results are not driven by outliers.
The validity of the results concerning H2, H3, and H4 was checked by ad-
dressing three statistical issues associated with regression analysis:
multicollinearity, heteroscedasticity, and influential points. The testing for
multicollinearity began by first calculating the correlation matrix of all the
independent variables in the regressions (see Table 6). All the coefficients were
found to be below the threshold level of 0.7 [75]. Two other popular diagnos-
tics were also calculated: the variance inflation factor (VIF) and the condition
number (CN). VIF was below the threshold level of 10 for the regression on
abnormal returns, and CN was below the threshold level of 30 [8, 19]. Thus,
multicollinearity was not a problem in the data set.
Because the data set includes cross-sectional data, variance of disturbances
may not be constant among all the data points because of different event-
specific or firm-specific factors related to data points. As a quick check, re-
siduals were plotted against independent variables, and no indication of trend
was found in the regression (i.e., the residuals did not increase or decrease
with the value of the independent variable). In addition, the White test was
performed [74]. Heteroscedasticity was not detected at the 5 percent level.
The question of whether results are driven by a set of influential observa-
tions is crucial. Removal of those observations may cause substantial changes
in the estimated coefficients and therefore in the test results. Cook’s distance
is the statistic usually employed to detect influential observations [22]. Sev-
eral researchers suggest that instead of using limits, a scree plot of Cook’s
Frequency of
positive
Event window Mean t-value p-valuea returnsc z-valueb p-valuea
0 0.007061 2.3671 0.010458 36 1.440512 0.07486

1 0.006540 2.0011 0.024780 35 1.357467 0.08732
0,1 0.013560 2.7123 0.004271 37 2.028216 0.02127
Table 9. Information-Transfer Effect on Security Firms.

Note: A total of 2,054 security firms were identified for 66 information-transfer events. The combination of the sample security firms varies with the timing of the
events, because the number of security firms that qualified for inclusion in the sample varies with time. The mean number of security firms per information-transfer
event was 31.12; the minimum was 17; the median was 36; and the maximum was 38.
a
p-values of one-tailed significance.
b
If T+ is the sum of the ranks assigned to positive CARs, and N is the sample size, then the test statistic is given by (T+ – a)/b, which is distributed as N(0,1) for large
samples, where a = N*(N + 1)/4, and b = N*(N + 1)*(2N + 1)/24.
c
Of 66 observations.
93
distances should be drawn to separate out the influential observations so that

they can be dropped [19, 21]. For the regression on cumulative abnormal re-
turns, the scree plot suggested the removal of five observations from the data
set. When the model was refitted without these points, firm type was still
significant at the 1 percent level, and attack type was still not significant.
The time variable in the regression analysis controlled for any change in
volatility through time, as suggested by earlier research. However, stock be-
tas can provide a firm-specific control for volatility. The regression was rerun
with betas included to control for firm-specific volatility in order to check
whether the results in Table 7 were confounded or not. Table 10 presents the
results of the new regression. Beta was significant at the 5 percent level. The
time variable that was a control variable became insignificant. This implies
that beta was a better control for volatility than time. However, the main re-
sults were still significant (firm type was even more significant than before).
The attack type was still insignificant. Therefore, the results of the cross-sec-
tional analysis were not confounded by stock betas. Thus, the results were
robust against every sensitivity and robustness check that was conducted.
Implications, Limitations, and Conclusions
Implications
According to the CSI-FBI Computer Crime and Security Survey 2002, which
polled 503 respondents from organizations throughout the United States, 80
percent reported financial losses, but only 44 percent (223) were able to quan-
tify them. The total reported loss was $455,848,000, and the average estimated
loss was $2,044,161 per organization across all types of breaches. The highest
reported loss was for theft of proprietary information, reported by 41 organi-
zations, with an average of $4,166,512 per organization. The sabotage of data
networks cost an average of $351,953, and DOS attacks resulted in a $244,940
loss per organization. The reported losses included the firms’ estimates of
direct and tangible costs associated with security breaches only.
The results of the present study show that the announcement of an Internet
security breach is negatively associated with the market value of the announc-
ing firm. In the study, breached firms lost an average of 2.1 percent of their
market value within the two days surrounding the events, which is roughly a
$1.65 billion average loss in market capitalization per incident across breach
types. This figure is orders of magnitude above the average loss estimate re-
ported in the CSI-FBI survey. The huge difference in the estimates of firm
losses because of a security breach may be explained by the fact that firms in
the CSI-FBI survey estimated only direct costs, such as lost productivity or
sales, and expenditure on restoring the breached system, whereas the loss
estimated in the present event study may also include investors’ expectations
about the impact on future cash flows, which requires considerations of intan-
gible costs, such as the loss of consumer confidence. Investors may also antici-
pate that the firm will be breached again in the future. The estimates based on
the event study may be noisy because of the uncertainties. Even if the esti-
Predicted Parameter Standard

sign estimate error t pa
Intercept –/+ –0.1389 0.0435 –3.19 0.0022

Firm type (H2) – –0.0539 0.0198 –2.72 0.0043
Firm size (H3) + 0.0131 0.0041 3.16 0.0013
Nature of attack (H4) –/+ –0.0126 0.0159 –0.79 0.4306
Time – –0.0053 0.0067 –0.79 0.2169
Beta –/+ 0.0416 0.0189 2.20 0.0318
Model R2 0.2902
Adjusted R2 0.2311
F-value 4.91 (0.0008)b
Table 10. Results of Cross-Sectional Regression by Firm Betas.

p represents one- (two-)tailed significance when a sign is (is not) hypothesized.

a
Number in parentheses specifies the significance of F.

b
mates are discounted, however, there is an order of magnitude of difference

between the firms’ reported estimates in the CSI-FBI survey and the market
value loss in the study. One possible implication of this finding is that the
intangible costs of security breaches may be much larger than the tangible
costs, and therefore, firms that ignore the intangible costs grossly underesti-
mate the loss from security breaches. Because investments in IT security are
directly dependent on the extent of potential loss from breaches, firms are
likely to underinvest in IT security if they make investment decisions based
only on tangible costs. The study offers proponents of IT security a much-
needed economic justification to get firms to invest in security.
The finding that the average (negative) cumulative abnormal return asso-
ciated with announcements decreases with the size of the firm suggests that
investors penalize smaller firms more than larger firms when a security breach
occurs. This implies that managers of small firms must understand the impor-
tance of security for their survival, and goes against the common argument,
“We are a small firm, and we cannot invest much in IT security.” In fact, man-
agers of small firms should view IT security as a key issue for survivability.
Another important finding is that the market penalizes all firms for secu-
rity breaches, but Internet firms are penalized more than conventional firms.
A possible explanation for this effect is the differential degree to which firms
depend on the Internet to generate revenues and survive in the long run. This
result is intuitive—firms that rely solely on the Internet for their revenue (so-
called pure-play firms) have more to lose from poor IT security infrastructure
than firms that have multiple sales channels. Security of IT systems is a matter
of the utmost importance for the success of Internet firms.
The results show that negative effects of security problems have been in-
creasing. A possible explanation for this progression is that the role of the
Internet in commerce has increased along with the risks. As investors come to
understand Internet-based business models, security breach events assume
more significance in their minds. An implication for managers is that the risk
of security breaches should be periodically reassessed, and the necessary steps
should be taken to mitigate the risk. The finding that firms experience similar
cumulative abnormal returns regardless of the nature of the attack is some-

what puzzling. One possible explanation is that investors regard any kind of
security breach as a failure of the IT security program and penalize firms for
not having taken adequate steps to prevent security problems. This explana-
tion implies that managers should focus on minimizing all types of breaches
to the fullest extent possible. Another possible explanation is that the intan-
gible costs associated with security breaches are quite difficult to estimate.
The large errors in these estimates obfuscate any differences between various
types of security breaches in investors’ minds. The study could not classify
attacks into more than two categories, because there were not enough obser-
vations. Additional insights require further investigation. To increase the power
of testing, future research should focus on more general classification of at-
tack types in a larger data set.

The finding that security firms realize significant positive returns as a reac-
tion to security breach announcements shows that security and e-commerce
go hand in hand. There can be no e-commerce unless there is adequate secu-
rity. The general rise in the stock prices of Internet security firms further justi-
fies this argument, because security can be improved only with a series of
security controls. Investors believe that firms will react to security breaches by
investing more in security technology. The study does not offer any insights
into what types of breaches will be prevented by particular technologies (e.g.,
firewalls, intrusion detection systems, disaster recovery systems). An effort to
answer this question would be an interesting extension of the study.
Limitations and Conclusions
The study and research design had several limitations. Event-study method-
ology was used to estimate the cost of security breach events for firms. This
methodology relies on the assumptions that markets are efficient and inves-
tors are rational. Thus, the present research, like all other event studies, suffers
from the fact that real markets can deviate from such an ideal characterization.
The time period used for the study was characterized by high market valua-
tion and market volatility. This by itself does not invalidate the event-study
methodology, but it may increase the errors in the estimates of breach cost.
Although it is difficult to quantify these effects, the study tried to control them
with the time variable.16 The time variable is only weakly significant. How-
ever, to the extent that time does not capture the effects mentioned above, the
results should be interpreted with caution. An effort was made to eliminate all
known confounding events from the sample, but other, unknown factors that
may affect a firm’s market valuation during the event window may increase
the error in the estimates of breach cost. Another limitation is the focus on
publicly traded U.S. firms. This limits the generalization of the results to other
countries but increases the internal validity of the research design. Future re-
searchers may study the issue with samples from other nations.
Another significant limitation of the study relates to the categorization of
breach types. Breaches were classified as either availability attacks or attacks
of some other type. Clearly, it would be beneficial to classify the breaches that
belong to the second category more precisely. The reason for the primitive
classification was the unavailability of data in the announcements. In addi-
tion, a finer classification would require more sample data to have sufficient
statistical power. Further research can provide additional insights into the
impact of breach type on capital markets.
The results of the study summarized in this paper should be reassuring to
firms that have invested in information security. They should, as well, be a
source of encouragement supporting implementation for firms that doubt the
value of adopting sound security practices.
NOTES
1. Subsequent to the study, Ettredge and Richardson extended their analysis to

include information-transfer effects associated with the February 2000 DOS attacks
[30].
2. Without loss of generality, the terminal period can be chosen to be infinity,
based on the going-concern assumption.
3. Subsequent work in this area has been done by Peteraf, Rubin, and
Wenderfelt [60, 65, 73].
4. Such attacks have been especially painful for AOL. Its customers have been
scammed several times using e-mail, telephone, and other means [44].
5. It is estimated that Internet traffic slowed by 26 percent because of DOS
attacks in February 2000 [45]. A study at the University of California at San Diego
estimated that nearly 4,000 such attacks per week are launched against Web sites
[24].
6. CloudNine Communications, one of Britain’s oldest Internet service provid-
ers, went out of business because its network was down for too long a time after a
DOS attack [72].
7. It is also possible that there may not be any adjustment in time if investors
have, on average, perfect insight into breach costs.
8. Note that the use of market-adjusted abnormal returns and time as a control
variable makes it possible to control for market- and time-specific volatility.
9. The market model is good in terms of detecting event effects. The variance of
abnormal return is reduced if the portion of the return related to variation in the
market’s return is removed. This improves the ability to detect event effects [46].
10. The use of other indices does not change our results qualitatively.
11. Unless there is a specific reason to keep the event window long, it should be
kept short. Using a long event window severely reduces the power of the test
statistics, which can cause false inferences about the significance of an event [11, 12].
Empirical evidence suggests that markets react to new information and adjust stock
prices almost instantaneously [27, 54]. Another reason to keep the event window
short is that it is almost impossible to control for confounding effects in long
windows, which, in turn, influences the robustness of the test results. The method-
ology herein assumes that abnormal returns are the result of the announcement,
and not of the other events occurring in the event window.
12. The cumulative abnormal return values are comparable to the cumulative
abnormal return values found in other event studies. For example, Bharadwaj and
Keil found that the mean cumulative abnormal return for IT failure announcements
was –1.79 percent [10]. Menzar, Nigh, and Kwok, who analyzed the effect of the
withdrawal of multinational firms from South African markets, observed that the
mean cumulative abnormal return was –2.0 percent [51].
13. The average market capitalization loss per incident was calculated as the
average market capitalization of breached firms times the average cumulative
abnormal return over the event window.
14. The choice of the first announcement date as the reference point does not
affect the results, nor does the choice of year as the time unit.
15. We thank Yexiao Xu, professor of finance at the University of Texas at Dallas,
for suggesting these procedures. They do not assume that the cross-sectional error
terms are independent for each event date. We also did a simple t-test of the mean
that assumes that the cross-sectional error terms are independent for each event
date. The result was highly significant (t = 4.95; p < 0.0001).
16. Researchers like Mikhail, Walther, and Willis have used this variable to
control for similar effects [52].
REFERENCES
1. Agrawal, D.; Bharath, S.T.; and Viswanathan, T. Technological change

and stock market volatility: Evidence from e-commerce adoptions. Working

Paper. University of Maryland at College Park, Robert H. Smith School of
Business, March 2003. Available at www.rhsmith.umd.edu/ceme/research/
TechnologicalChange-April14-2003.pdf.
2. Angus Reid Group. Security and Privacy Issues Keeping Millions from
Shopping Online. Angus Reid Group, April 27, 2000 (www.ipsos-na.com/
news/pdf/media/ap000426.pdf).
3. Asthana, S.C., and Mishra, B.K. The differential information hypothesis,
firm size, and earnings information transfer: An empirical investigation.
Journal of Business Research, 53 (2001), 37–47.
4. Atomic Tangerine. NPV: Information Security. Atomic Tangerine, Pitts-
burgh, 2000 (www.ttivanguard.com/risk/netpresentvalue.pdf).
5. Baginski, S.P. Information transfer associated with management fore-
casts of earnings. Journal of Accounting Research, 25 (1987), 196–216.
6. Banz, R. The relationship between return and market value of common
stocks. Journal of Financial Economics, 9 (March 1981), 3–18.
7. Beale, M. Survey reveals consumer concern over e-commerce security
issues. E-Commerce Times, June 21, 1999 (www.ecommercetimes.com/story/
1981.html).
8. Belsley, D.; Kuh, E.; and Welsch, R. Regression Diagnostics. New York:
John Wiley, 1980.
9. Bernard, V.L. Cross-sectional dependence and problems in inference in
market-based accounting research. Journal of Accounting Research, 25, 1
(1987), 1–48.
10. Bharadwaj, A., and Keil, M. The effect of information technology failures
on the market value of firms: An empirical examination. INFORMS 2001
Miami, November 2001 (www.informs.org/Conf/Miami2001/TALKS/
Sponsor-13.html).
11. Brown, S., and Warner, J. Measuring security price performance. Journal
of Financial Economics, 8 (1980), 205–258.
12. Brown, S., and Warner, J. Using daily stock returns: The case of event
studies. Journal of Financial Economics, 14 (1985), 3–31.
13. Campbell, J.Y.; Lo, A.W.; and MacKinlay, A.C. The Econometrics of Finan-
cial Markets. Princeton: Princeton University Press, 1997.
14. Caves, R.E. Industrial organization, corporate strategy and structure.
Journal of Economic Literature, 58 (1980), 64–92.
15. CBS News.com. Locking Windows. Associated Press, January 16, 2003.
Available at www.cbsnews.com/stories/2002/01/16/tech/
main324663.shtml.
16. CERT Coordination Center. CERT/CC Statistics 1988–2003. Carnegie-
Mellon University, Software Engineering Institute, January 22, 2004. Avail-
able at www.cert.org/stats/cert_stats.html.
17. Chang, H.; Mishra, B.; and Huang, C. Manager’s earnings forecast and
information transfer in supply chains. Working Paper, University of Califor-
nia at Riverside, 2003.
18. Chatterjee, D.; Richardson, V.J.; and Zmud, R.W. Examining the share-
holder wealth effects of announcements of newly created CIO positions.
MIS Quarterly, 25, 1 (2001), 43–70.
19. Chatterjee, S., and Price, B. Regression Analysis by Example. New York:
John Wiley, 1991.
20. Clinch, G.J., and Sinclair, N.A. Intra-industry information releases: A
recursive system approach. Journal of Accounting and Economics, 9 (1987), 89–
106.
21. Coenders, G., and Saez, M. Collinearity, heteroscedasticity and outlier
diagnostics in regression. In A. Ferligoj and A. Mrvar (ed.), New approaches
in applied statistics. Ljubljana: FDV, 2000, pp. 79–94.
22. Cook, R.D. Detection of influential observations in linear regression.
Technometrics, 19 (1977), 15–18.
23. Cooper, M.J.; Dimitrov, O.; and Rau, P.R. A rose.com by any other name.
Journal of Finance, 56, 6 (December 2001).
24. Costello, S. Study: Nearly 4000 DOS attacks occur per week. CNN.com/
IDG.net. Available at www.cnn.com/2001/TECH/internet/05/24/dos
.study.idg/.
25. CSC (Computer Sciences Corporation). CSC survey reveals inadequate
information security practices among companies worldwide (November 19,
2001). Available at www.csc.com/newsandevents/news/1584.shtml.
26. D’Amico, A.D. What does a computer security breach really cost? Secure
Decisions, a Division of Applied Visions, Northport, NY, September 7, 2000.
27. Dann, L.; Mayers, D.; and Raab, R. Trading rules, large blocks and the
speed of price adjustment. Journal of Financial Economics, 4 (1977), 3–22.
28. Davidson, W.; Worrell, D.; and Dutia, D. The stock market effects of CEO
succession in bankrupt firms. Journal of Management, 16 (1993), 517–533.
29. Dos Santos, B.L.; Peffers, D.C.; and Mauer D.C. The impact of informa-
tion technology investment announcements on the market value of the firm.
Information Systems Research, 4, 1 (1993), 1–23.
30. Ettredge, M., and Richardson, V.J. Assessing the risk in e-commerce. In
R.H. Sprague, Jr. (ed.), Proceedings of the Thirty-fifth Hawaii International
Conference on System Sciences. Los Alamitos, CA: IEEE Computer Society
Press, 2002.
31. Fama, E. Efficient capital markets II. Journal of Finance, 46, 5 (1991), 1575–1617.
32. Fama, E.; Fisher, L.; Jensen, M.C.; and Roll, R. The adjustment of stock
prices to new information. International Economic Review, 10, 1 (1969), 1–21.
33. Fama, E., and French, K. The cross-section of expected stock returns.
Journal of Finance, 47, 2 (1992), 427–465.
34. Foster, G. Intra-industry information transfer associated with earnings

releases. Journal of Accounting and Economics, 3 (1981), 201–231.
35. Freeman, R., and Tse, S. An earnings prediction approach to examining
intercompany information transfers. Journal of Accounting and Economics, 15
(1992), 509–523.
36. Friedman, S.D., and Singh, H. CEO succession and stockholder reaction:
The influence of organizational context and event context. Academy of
Management Journal, 32 (1989), 718–744.
37. Han, J.C.Y.; Wild, J.J.; and Ramesh, K. Managers earnings forecast and
intra-industry information transfers. Journal of Accounting and Economics, 11
(1989), 1–33.
38. Hand, J. Profits, losses, and non-linear pricing of internet stocks. In J.
Hand and B. Lev (eds.), Intangible Assets: Values, Measures, and Risks. New
York: Oxford University Press, 2003.
39. Hendricks, K.B., and Singhal, V.K. The effect of supply chain glitches on
shareholder wealth. Paper presented at INFORMS 2001, Miami, November 2001.
40. Im, K.S.; Dow, K.E.; and Grover, V. Research report: A reexamination of
IT investment and the market value of the firm: An event-study methodol-
ogy. Information Systems Research, 12, 1 (March 2001), 103–117.
41. Information Security. Information Security 2002 Excellence Awards
Listing of Categories and Semi-Finalists. Available at www.infosecuritymag
.com/awards2002/listproducts.html.
42. INFOSYSSEC, The Security Portal for Information System Security
(http://infosyssec.com).
43. Koh, J., and Venkatraman, N. Joint venture formations and stock market
reactions: An assessment in information technology sector. Academy of
Management Journal, 34 (1991), 869–892.
44. Kornblum, J. AOL users conned for credit data. CNET News.com, May
20, 1997. Available at http://news.com.com/2100–1033_3–279948.html.
45. Lemos, R. A year later, DOS attacks still a major Web threat. CNET
News.com, February 7, 2001. Available at http://news.com.com/
A+year+later percent2C+DDoS+attacks+still+a+major+Web+threat/2009-
1001_3-252187.html.
46. MacKinlay, C.A. Event studies in economics and finance. Journal of
Economic Literature, 35 (1997), 13–39.
47. Mahdawan, R., and Prescott, J.E. Market value impact of joint venture:
The effect of industry information-processing load. Academy of Management
Journal, 38 (1995), 900–915.
48. Maiwald, E. Network Security: A Beginner’s Guide. Berkeley: Osborne/
McGraw-Hill, 2001.
49. Markoff, J. Stung by security flaws, Microsoft makes software safety a
top goal. New York Times, January 17, 2002, C1.
50. McWilliams, A., and Siegel, D. Event studies in management research:
Theoretical and empirical issues. Academy of Management Journal, 40, 3
(1997), 626–657.
51. Menzar, M.B.; Nigh, D.; and Kwok, C.C.Y. Effect of announcements of
withdrawal from South Africa on stockholder wealth. Academy of Manage-
ment Journal, 37, 6 (1994), 1633–1648.
52. Mikhail, M.; Walther, B.; and Willis, R. Do security analysts improve
their performance with experience? Journal of Accounting Research, 35
(Suppl.) (1997), 131–157.
53. Miller, I., and Miller, M. John E. Freund’s Mathematical Statistics (6th ed.).
Upper Saddle River, NJ: Prentice Hall, 1999.
54. Mitchell, M., and Netter, J. Triggering the 1987 stock market crash:
Antitakeover provisions in the proposed house ways and means tax bill?
Journal of Financial Economics, 24 (1989), 37–68.
55. Nersey, C. Hacker scare again boosts security stocks. Available at
www.internetnews.com/bus-news/article.php/573961.
56. Olsen, C., and Dietrich, R. Vertical information transfers: The association
between retailers’ sales announcements and suppliers’ security returns.
Journal of Accounting Research, 23 (1985), 144–166.

57. Otchere, I. Two decades of information transfer studies: A review.
Accounting Research Journal, 15, 1 (2002), 21–38.
58. Pastore, M. Companies lack understanding of information security.
ClickZ, October 10, 2001. Available at http://cyberatlas.internet.com/
big_picture/hardware/article/0,1323,5921_900911,00.html.
59. Penrose, E.T. The Theory of the Growth of the Firm. New York: John Wiley,
1959.
60. Peteraf, M.A. The cornerstones of competitive advantage: A resource-
based view. Strategic Management Journal, 14 (1984), 179–191.
61. Power, R. 2002 CSI/FBI computer crime and security survey. Computer
Security Issues and Trends, 8, 1 (2002).
62. Rajgopal, S.; Venkatachalam, M.; and Kotha, S. Managerial actions, stock
returns, and earnings: The case of business-to-business internet firms.
Journal of Accounting Research, 40, 2 (May 2002).
63. Reinganum, M.R. Misspecification of capital asset pricing: Empirical
anomalies based on earnings yield and market values. Journal of Financial
Economics, 9 (March 1981), 19–46.
64. Rombel, A. Internet security in an insecure world. Global Finance, 15, 13
(December 2001), 28–32.
65. Rubin, P.H. The expansion of firms. Journal of Political Economy, 81 (1973),
936–949.
66. Schipper, K. Information transfers. Accounting Horizons, 4 (1990), 94–107.
67. Seth, A. Value creations in acquisitions: A reexamination of performance
issues. Strategic Management Journal, 11 (1990), 99–115.
68. Shelton, L. Strategic business fits and corporate acquisition: Empirical
evidence. Strategic Management Journal, 9 (1988), 278–288.
69. Subramani, M., and Walden, E. The game of the name: A comparison of
capital market reactions to dotcom vs. traditional name changes. MISRC
Working Paper No. 00-15. University of Minnesota, Carlson School of
Management, July 18, 2001. Available at http://misrc.umn.edu/
workingpapers/fullPapers/2000/0015_071800.pdf.
70. Subramani, M., and Walden, E. The impact of e-commerce announce-
ments on the market value of firms. Information Systems Research, 12, 2 (June
2001), 135–154.
71. Trueman, B.; Wong, M.H.; and Zhang, X.J. The eyeballs have it: Search-
ing for the value in internet stocks. Journal of Accounting Research, 38 (2000),
137–162.
72. Warner, B. Internet firm hacked out of business. Reuters, February 1,
2002. Available at http://techupdate.zdnet.com/techupdate/stories/main/
0,14179,2844881,00.html.
73. Wenderfelt, B. A resource based view of the firm. Strategic Management
Journal, 5 (1984), 171–180.
74. White, H. A heteroscedasticity-consistent covariance matrix estimator
and a direct test for heteroscedasticity. Econometrica, 48, 4 (1980), 817–838.
75. Zhu, K., and Kraemer, K.L. E-commerce metrics for net-enhanced
organizations: Assessing the value of e-commerce to firm performance in
the manufacturing sector. Information Systems Research, 13, 3 (2002), 275–295.
Appendix A: Illustrative Samples of Internet Security

Breach Announcements
Example of a Single Security Breach Announcement
Source: CNet News.Com

Time: January 10, 2000
Title: FBI Probes Extortion Case at CD Store
Taking advantage of an apparent security breach in CD Universe’s data-
base, a hacker posted links to potentially thousands of customer names,
addresses, and credit card numbers after purportedly failing to extort
money from the on-line music store. The hacker, who goes by the name
“Maxus,” claimed to have tapped into an estimated 350,000 user names
and credit cards from CD Universe. Before posting access to the infor-
mation through a Web site identified only by its IP address, he demanded
$100,000 from the Connecticut-based Web store. In a statement issued
today, eUniverse, which owns CD Universe, acknowledged that a por-
tion of its customer data had been stolen. The company said it had noti-
fied the FBI after the hacker attempted blackmail. The FBI shut Maxus’
Web site Saturday after it was discovered, the company said.
Example of a Multiple Security Breach Announcement
Source: CNet News.Com

Time: February 15, 2001
Title: Hackers Attack HP, Compaq, Others
A group of Internet vandals calling themselves Sm0ked Crew has hit the
Web sites of technology giants such as Hewlett-Packard, Compaq Com-
puter, Gateway. Late Wednesday, Attrition.org, an independent organi-
zation that records hacking incidents, reported that Sm0ked Crew had
defaced HP’s e-learning site, a Compaq Europe, Middle East, and Africa
page. Wednesday morning, the affected HP and Compaq sites still car-
ried a message from the hackers: “Admin, You just got Sm0ked. This site
was hacked by Sm0ked crew. Hacked by splurge and The-Rev. Greetz
dislexik, nouse, system33r, italguy, B_Real and anyone I forgot,

sm0kedcrew@hushmail.com.”
Appendix B: List of Internet Security Firms
This is a list of security firms traded in the United States. The list was com-
piled from two sources: INFOSYSSEC, the security portal for information sys-
tem security, and Information Security Magazine, the leading magazine of the
security industry. A careful investigation of the firms on the list found that IT
security was not the core business for some of them. The business description
of each firm was checked in Yahoo Finance. If security was not mentioned
there, the firm was dropped from the list. Sixteen firms were dropped because
their core businesses did not include security. An asterisk indicates that the
firm was dropped from the list.
ID Company ID Company
1* Hewlett Packard Co. 29* Lucent Technologies Inc.

2* International Business Machines Co. 30 Axent Technologies Inc.
3* National Service Industries Inc. 31* S1 Co.
4* Schlumberger Ltd. 32* Electronic Data Systems Co.
5* Unisys Co. 33 Check Point Software Tech Ltd.
6 Computer Associates Intl Inc. 34 V -One Co.
7 Datakey Inc. 35 Sac Technologies Inc.
8* 3COM Co. 36 Verisign Inc.
9* Novell Inc. 37 Internet Security Systems Inc.
10* Sun Microsystems Inc. 38 Bindview Development Co.
11* Microsoft Co. 39 Pilot Network Services Inc.
12 Rainbow Technologies Inc. 40 Entrust Technologies Inc.
13 Netegrity Inc. 41 Network1 Security Solutions Inc.
14 Symantec Co. 42 Hifn Inc.
15 ZixIt Co. 43 Trend Micro Inc.
16* Cisco Systems Inc. 44 Netiq Co.
17 Identix Inc. 45 Watchguard Technologies Inc.
18* Bmc Software Inc. 46 Tumbleweed Communications Co.
19 Intrusion Com Inc. 47* Red Hat Inc.
20 Networks Associates Inc. 48 Baltimore Technologies Pl
21 Safenet Inc. 49 Predictive Systems Inc.
22 Saflink Co. 50* Akamai Technologies Inc.
23 Wave Systems Co. 51 Sonicwall Inc.
24 Cyberguard Co. 52 Mcafee Com Inc.
25 Rsa Security Inc. 53 ActivCard S.A.
26 Secure Computing Co. 54 Vasco Data Sec Intl Inc.
27 Aladdin Knowledge Systems 55 Gemplus International S.A
28 Cylink Co. 56 Jawz Inc.
HUSEYIN CAVUSOGLU (huseyin@tulane.edu) is an assistant professor of informa-

tion and operations management at the A.B. Freeman School of Business at Tulane
University. He received his Ph.D. in management science with a specialization in MIS
from the University of Texas at Dallas. He has presented his work at the International
Conference on Information Systems (ICIS), Workshop on Information Systems and
Economics (WISE), Workshop on Information Technology and Systems (WITS), Ameri-
cas Conference on Information Systems (AMCIS), and other conferences, and has pa-
pers forthcoming in Communications of the ACM and Decision Analysis. His major re-
search interests are in the areas of information economics, assessment of the value of
IT security, and IT security management. He is a member of AIS and INFORMS.
BIRENDRA MISHRA (barry.mishra@ucr.edu) is an assistant professor of accounting

in the School of Management at the University of California, Riverside. He received
his Ph.D. in business with a specialization in accounting from the University of Texas
at Austin. His research areas include management accounting and management con-
trol issues, disclosure, policy-making and regulatory issues, information sharing in
supply chains, and IT security. He has published in Journal of Accounting Research,
Marketing Science, Management Science, Journal of Accounting and Public Policy, Journal of
Management Accounting Research, Journal of Business Research, and American Institute of
Chemical Engineers Journal.
SRINIVASAN RAGHUNATHAN (sraghu@utdallas.edu) is an associate professor of

management information systems in the School of Management at the University of
Texas at Dallas. He received his Ph.D. in business from the University of Pittsburgh.
His research interests concern the economics of information systems. He has pub-
lished in Management Science, Information Systems Research, Journal of Management In-
formation Systems, and IEEE Transactions.

The Effect of Internet Security Breach Announcements On Market Value: Capital Market Reactions For Breached Firms and Internet Security Developers

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Effect of Internet Security Breach Announcements On Market Value: Capital Market Reactions For Breached Firms and Internet Security Developers

Uploaded by

Copyright:

Available Formats

International Journal of Electronic Commerce

ISSN: 1086-4415 (Print) 1557-9301 (Online) Journal homepage: http://www.tandfonline.com/loi/mjec20

The Effect of Internet Security Breach

Huseyin Cavusoglu Ph.D. , Birendra Mishra Ph.D. & Srinivasan Raghunathan

To link to this article: http://dx.doi.org/10.1080/10864415.2004.11044320

Published online: 08 Dec 2014.

Submit your article to this journal

View related articles

Citing articles: 4 View citing articles

Full Terms & Conditions of access and use can be found at

ABSTRACT: Assessing the value of information technology (IT) security is challenging

An earlier version of this paper was presented at WISE 2002 (Workshop on

The research in this paper is related to two previous studies of IT failures

Tangible Low High

Table 1. Degree of Uncertainty in Estimation of Costs.

Theory and Hypotheses Development

Cost of Security Breaches

Investors’ Beliefs and Firm Values Under an Efficient Market

∆Vb = ∆V −∆Vm . (3)

Impact of Security Breaches on Market Value of Breached Firm

Determinants of Cross-Sectional Variance

The discussion in this section develops hypotheses to explain the cross-sec-

three categories: firm-specific factors, context-specific factors, and event-spe-

Relationship Between Internet Security Breaches and Market Value of

Announcements of security breaches convey information to investors about

current 0.4 percent of revenue to 4 percent of revenue by 2011, a 1,000 percent

Data Set and Methodology

For the purpose of analyzing how announcements of security breaches affect

Web defacement (see Appendix A for an illustrative sample). Consistent with

not publicly traded—government agencies, nonprofit organizations, and pri-

sistencies in categorization were resolved through discussions. Of the 66 events,

Table 2 presents the distribution of security breach announcements in the fi-

Selection of Internet Security Product Firms

Table 2. Security Breach Announcements, 1996–2001.

The event-study methodology was used to assess the impact of announce-

Total assets (in million $) 38,558.9 5,348.0 109,158.2 37.8 642,191.0

Table 3. Descriptive Statistics for Breached Firms.

The event of interest in the present study is the announcement of an Internet

These are aggregated across all events to draw an overall inference as

Mean –0.0086 –0.0123 –0.0209

Table 4. Descriptive Statistics of Abnormal Returns and Cumulative

Effect of Security Breach Announcements on Announcing

0 –0.008638 –1.74985 0.04243 36 –0.948630 0.171404

Table 5. Event Study Results for Breached Firms.

Determinants of Cross-Sectional Variance in Cumulative

CARi = β 0 + β1 ( FirmType )i + β 2 ( FirmSize )i + β 3 ( AttackType )i

The independent variables were operationalized as follows. Using the cod-

(1) Firm type 0.47 N/A 0 1 1.00

Table 6. Descriptive Statistics and Pearson Product–Moment Correlations.

Predicted Parameter Standard

Intercept –/+ –0.1141 0.0433 –2.64 0.0106

Table 7. Results of Cross-Sectional Regression to Predict Variability in

p represents one- (two-)tailed significance when sign is (is not) hypothesized.

Number in parentheses specifies the significance of F.

Effect of Security Breach Announcements on

because of the presence of positive industry cross-sectional return covariation

Statistics t=0 t=1 (t = 0, t = 1)

Mean 0.0071 0.0065 0.0136

Table 8. Descriptive Statistics of Abnormal Returns and Cumulative

Sensitivity Analysis of Results

The robustness of the results to influential observations and model assump-

0 0.007061 2.3671 0.010458 36 1.440512 0.07486

Table 9. Information-Transfer Effect on Security Firms.